MILU
Multi-task Indic Language Understanding Benchmark
Updated
Cite
Fill in the access date before citing.
Plain text
Benchmark Registry. Multi-task Indic Language Understanding Benchmark Original benchmark results. Benchmark identifier: milu/original. https://benchmarkregistry.org/benchmarks/milu/original. Accessed [YYYY-MM-DD]. Data license: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Attribution: Benchmark Registry, a project of Densa Labs — https://benchmarkregistry.org/ (CC BY 4.0). Include the cited record URL and identify any changes.
BibTeX
@misc{br_milu_original,
title = {Multi-task Indic Language Understanding Benchmark Original benchmark results},
publisher = {Benchmark Registry},
url = {https://benchmarkregistry.org/benchmarks/milu/original},
urldate = {YYYY-MM-DD},
note = {Benchmark identifier: milu/original; data license: CC BY 4.0; access date to be filled by the reader}
}Scores over time
Best recorded score per model; higher is better. The line follows the running best. Evaluation contexts can differ.
Data table
| Model | Date | Score |
|---|---|---|
| Claude Opus 4.7 | (Reported) | 89.9% |
| Claude Sonnet 5 | (Reported) | 89.3% |
| Claude Opus 5 | (Reported) | 92.1% |
| Claude Fable 5.1 | (Reported) | 93.0% |
| Claude Opus 5.5 | (Reported) | 93.1% |
| Claude Sonnet 5.5 | (Reported) | 91.6% |
| Claude Haiku 5.5 | (Reported) | 87.6% |
Results
7 results